[Decision] #15484's ruling says packages/rest/vitest.config.ts opts the suite down — executing that sentence reddens 28 assertions and turns 8 files' "logs NOTHING" pins vacuous

维护者速读

事情 —— 批 #49 裁 #15484 时写了一句:「packages/rest/vitest.config.ts 把这个套件调低」。座位把它派出去实施,dev 在动手前先量了一遍,结果是这句话照字面执行会伤到门禁本身:

把套件调到 silent,栈帧确实从 2,095 → 0(输出少 36.7%);
但15 个文件里 28 条断言当场变红——它们断言的是「运维还是拿得到那句话」,而且是通过 console.error 的 mock 读的,本来一行量都没贡献;
更糟的一半:8 个文件断言的是反面——「一个预期内的 4xx 一行都不该记」。套件被静音之后,这些断言会因为错误的原因变绿:就算每个 4xx 都在狂刷日志,它们照样绿。

⚠️ 这正是这张卡自己踩过的坑:它原来的对照词 [Registry] 就是被 #15425 一条完全正当的声明悄悄作废的,于是「对照归零」看起来像「抓取失败」。⇒ 把套件静音、其它什么都不改,等于用一条看起来正当的声明把一道门禁变成摆设。

已经落地的部分不受这个问题影响:声明式 seam(OS_REST_LOG)、门禁脚本、README、审计文档的裁决交接,都在 PR #17863 里,出厂默认一字未动('info',行为与今天逐字节相同),而且默认值本身被门禁钉住——调到 error 或 silent 会让门禁红。⇒ 本卡只问剩下的那一件:套件的级别怎么设。

四个选项(卡上有完整四维分析)

A —— 维持现状:套件仍然响亮,2,095 行栈帧照旧。⛔ 代价是裁决那句话没兑现,量到的 +50.7%/窗口的增长没人管。
B —— 调低套件,同时给那 ~15 个断言文件各自显式声明响亮级别,并加一道配对门禁:一个 spy 了 console.error 却没声明级别的文件算 finding。⛔ 代价是 ~20 个文件 + 第二道门禁,而且「防摆设」这件事只有那道配对门禁真落地才算数。
C —— 调低,其它什么都不改。⛔ 28 条红 + 8 个文件变摆设,列在这里只是为了记录它被量过并被否掉,不是被忽略。
D —— 明确把套件的音量问题按「就该这么响」关掉,seam 只留给运维。⛔ 这与裁决自己那句话相反,只有维护者能改。

⇒ 请裁一个字母:A、B、C 还是 D?

What this card is, in one paragraph

Decision batch #49 (5551137024) ruled option A on #15484 and its execution list included one sentence: 「packages/rest/vitest.config.ts opts the suite down」. PR #17863 delivers every other part of that ruling — the declared OS_REST_LOG seam on logError, a gate that reads it (scripts/check-rest-log-declared.mjs, 19-case self-test, wired into package.json and lint.yml), the published README section, and the audit's lines 419–420 discharge — with the shipped default unchanged and gate-pinned. This card carries only the opt-down sentence, because executing it as written is a larger and riskier act than the ruling's author could have known when writing it.

⭐ This is not a re-litigation of the ruling. 「A ruling is not settled for a case until its instruction is executable ON that case」 — and the measurement below is what makes that determination, not an opinion about the ruling.

The measurement, taken before anything was chosen

Taken by the implementing dev on the delivered branch, with the control validated in the same capture it was used in ([sql-driver] DATABASE_ERROR = 362 and [REST] = 272 both fire; [Registry] reads 0, as this card's own thread documented at 5550786137):

reading	value
captured lines, one green `packages/rest` run	5,709
indented `at` stack frames	2,095 (36.7%)
frames arriving through `logError`	100%
frames with `OS_REST_LOG: 'silent'`	0
assertions that go RED at `'silent'`	28, across 15 files
files whose "an expected 4xx logs NOTHING" assertions become vacuous	8 (4 such assertions in `rest-expected-error-logging.test.ts` alone)

The 28 are not the four pins this card already knew about (#5437 / #4886 / #5489) — they are 7× that, and they read the fault through a console.error mock, so they never printed any of the volume being complained about.

Four-axis analysis — 四维

① 项目长远合理性(权重恒 ≥50%,领起) —— 维护者 2026-09-01 裁,逐字:「四维分析中,长期合理应该权重最高,至少50%」。

会烂掉的是声明和默认值,不是音量——而这两样已经被门禁焊住了(默认值降到 error/silent 直接门禁红)。⇒ A 已经买到了耐久的那一半。B 的机器结构本身是一项长期负债:一个「默认不记故障日志」的套件,再由每个文件自己重新调响,只有在那道配对门禁同时落地时才是安全的;缺了它,B 比 A 更糟。D 把一句裁决直接反掉,只能由维护者来做。C 长期上就是把一道门禁换成摆设。

② 真实业务需要 —— 每次跑 2,095 行、占输出 36.7%,并且这张卡量到的是增长(窗口内 +50.7%,而 driver 那一支只有 +2.0%)。真实,但它是摩擦不是缺陷,也不在任何用户路径上。⇒ 支持「要做」,不支持「现在连着做」。

③ 防 AI 犯错 —— 决定性的一轴。B 若没有那道配对门禁,就留下一个陷阱:在一个被静音的套件里断言「什么都没记」永远是绿的。这张卡自己就是被同一形状咬过的——[Registry] 对照被 #15425 一条正当声明悄悄作废。⇒ B ⛔ 不能顺手落;而先落 A 不花任何代价,也不堵死任何一条路(套件级别是 vitest.config.ts 里的一个字符串,B 的门禁到位后一行就能翻过去)。

④ 防过度设计 / 防范围蔓延 —— B 是派发令没有点名的 ~20 个文件,外加第二道针对测试文件的门禁。派发令自己的 stop-and-report 条件说的就是这种情况,dev 停下来是对的。⇒ A 现在,B 单独立卡。

⇒ 四轴同向 A(本 PR)+ B 独立立卡。⚠️ 但 A 与裁决自己那句话不一致,所以 ⛔ 不走代裁:这需要维护者一个字母。

Dedup, with the bound stated

Enumerated 534 open issues (paged REST, complete) and grepped titles + full bodies: OS_REST_LOG → 0, opt[- ]down → 0, packages/rest/vitest\.config → 0. ⚠️ Control OS_REGISTRY_LOG → 0 as well (a dead control on this population), so the reading rests on the live one: logError → 1, and that one hit is #15484 itself ⇒ the body matcher is alive. #15484's own 10-comment thread was read end to end; the opt-down appears there only inside the ruling's execution list and the dev's report, never as a card.

⛔ Ungraded and unrouted on purpose — domain:*, type and the priority are triage's to set, and this seat does not set them.

⛔ This card holds nothing and waits on nothing mechanical — PR #17863 does not depend on the answer and ⛔ must not be held for it. (⚠️ No Blocked-by key is written here on purpose: that marker is line-initial and decoration-tolerant, so writing it with a prose target would create exactly the half-state this board repairs.)

Generated by Claude Code

Ruling recorded — B: the `packages/rest` suite is opted down, every test file that asserts on fault-log output declares its own level, and a pairing gate makes an undeclared `console.error` spy a finding — all in one delivery, ⛔ never the opt-down alone (director seat, decision batch #128 item 4, 2026-09-13)

Maintainer, verbatim (live PM chat, reply received 2026-09-13T06:3xZ), to decision batch #128 presented as 1(a)·2B·3(4)·4B·5(1′): 「17469 从业务的角度给我具体解释。其他同意」 — 「其他同意」 covers this card.

Derived first from the long-term axis: a test that asserts on logging behaviour must own the log level it runs under; a suite whose ambient level decides whether "logs nothing" is a real assertion or a vacuous one is a suite that can go blind on a legitimate-looking config change — this card's own [Registry] control was silenced exactly that way by #15425. B is the only shape under which that rule holds and stays held (the pairing gate). A and D leave 2,095 stack-frame lines (36.7% of the output) on every green run, which is noise a human or an agent reading CI output has to wade through — an error source of its own. C was measured and refused by the card. ②③④ do not overturn ①; ④ wanted "A now, B as its own card", which is C for the interval, and is not taken.

What is ruled

packages/rest/vitest.config.ts sets the suite's OS_REST_LOG to the quiet level — executing batch Add granular query operation capabilities to driver schema #49's sentence (5551137024) as it was meant.
The ~15 files whose 28 assertions read the fault line through a console.error mock declare the loud level in their own setup; the 8 files asserting "an expected 4xx logs NOTHING" do the same, so their negative assertions keep meaning something.
A pairing gate (scripts/, wired like check-rest-log-declared.mjs, with its own self-test): a test file that spies on console.error / the fault logger without declaring OS_REST_LOG is a finding by name. It lands in the same PR as 1; ⛔ 1 does not merge without 3.
The shipped default stays 'info' and gate-pinned (PR feat(rest): a declared OS_REST_LOG fault-log level seam, shipped default unchanged #17863); nothing in this ruling touches production behaviour.

State

needs-user-decision → pm:queue; domain:cli / priority:p3 kept. Cross-domain: designated lane domain:cli (the test files), the scripts/ gate declared in the claim comment per the cross-domain exception path. Clause-②: no. #15484's execution list reads this comment as the authority for its opt-down sentence.

Generated by Claude Code

Claim: PM loop round R74
Session: session_01TbSMtGzMrtPwh925wDEZd5
Branch: claude/issue-17865-rest-log-suite-opt-down
Worktree: objectstack-issue-17865
Domain: domain:cli
File surface: packages/rest/vitest.config.ts; the packages/rest/src/** test files that assert on fault-log output; a new pairing gate under scripts/ with its own self-test; and its wiring in the root package.json and .github/workflows/lint.yml (stop on breach; explain in the report)
Container & model: M, mode:subagent, model: default judgment tier — dispatch-gates.mjs --tier returned no path-derived mandate on this surface, so the tier is this seat's call under standing ruling 5612096863 (floor sonnet · default opus · ceiling fable)
Clause-②: no
Thread-read: 5651668684
Serial constraints cleared: see the targeted in-flight check below — no in-flight card in either affected domain declares an intersecting surface. ⚠️ The gate wiring touches the shared root package.json and .github/workflows/lint.yml; declared here so any concurrent seat can see it. Latest origin/main touch on this card's own surface is 3a5eaea54 (#17863, the seam this card completes).

Cross-domain exception path — claimed under triage's own designation, ⛔ not this seat's improvisation

This card's file surface spans two domains: packages/rest/** (domain:cli) and scripts/ gate-class (domain:spec under the anchoring-rule exception). The designation is already recorded by the director seat in the ruling comment 5651668684:

Cross-domain: designated lane domain:cli (the test files), the scripts/ gate declared in the claim comment per the cross-domain exception path.

⇒ 跨域例外路径 requires the lane be designated by triage and the full surface declared in the claim. Both are satisfied; the declaration is the File surface line above.

Targeted in-flight check — run because this IS the cross-domain path

范围 = the domains this card's surface touches, ⛔ not a global sweep. Read from each in-flight card's own claim comment:

in-flight card	lane	declared surface	intersects?
#18070	`domain:cli`	`packages/client/src` + two `packages/runtime/src/*.integration.test.ts`	⛔ no
#18078	`domain:spec`	none — report-only, changes no file and opens no PR	⛔ no, by construction
#16929	`domain:spec`	⚠️ newest claim (2026-09-12T08:31:43Z, `os-bill`) declares a branch but no `File surface:` line this seat can read	⚠️ see below

⚠️ #16929's non-declaration is stated rather than smoothed. It is an assignedProfiles property retirement, whose work lands in packages/spec/src plus generated baselines, forms and pin tests — none of which is on this card's surface. But that is this seat's inference from the card's subject, ⛔ not a reading of a declaration, because there is no declaration to read. The one genuine contact risk is the shared gate wiring (root package.json, lint.yml), which is additive here (one script entry) and which the merge queue arbitrates. ⛔ Not treated as cleared by silence; treated as low and declared.

Premise re-verified against the TREE, ⛔ not against the card

origin/main = fb29f62ce. ⚠️ This card's body was written before its ruling and argues for option A; the ruling chose B. The body is therefore stale on the decision and the ruling comment is the authority — see the dispatch order.

premise	reading
the `OS_REST_LOG` seam landed	✅ `3a5eaea54` (#17863); `OS_REST_LOG` resolves in 7 files repo-wide
the gate-wiring precedent exists	✅ `scripts/check-rest-log-declared.mjs` present
ruling item 1 is still undone	✅ `packages/rest/vitest.config.ts:35` and `:55` both still read `OS_REST_LOG: 'info'` — the suite is still loud

⚠️ One instrument bound declared: git grep -l "console.error" -- packages/rest/src returns 4 files, while the card's population is "~15 files / 28 assertions". These are ⛔ not the same measurement — the card counts assertions reading the fault line, which need not go through a literal console.error spy. 4 is a floor, ⛔ not a contradiction of 15, and the delivering dev re-measures the real population rather than inheriting either number.

_{domain:cli execution PM seat · #6024 · session session_01TbSMtGzMrtPwh925wDEZd5 · R74 · 2026-09-13T23:3xZ}

Generated by Claude Code

os-dev-report

{
  "issue": 17865,
  "status": "done",
  "branch": "claude/issue-17865-rest-log-suite-opt-down",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/18090",
  "premise_still_valid": true,
  "head": "6d699ac6f",
  "base": "c54d8d67b",
  "ruling_items_delivered": {
    "1": "packages/rest/vitest.config.ts OS_REST_LOG 'info' -> 'silent' in all 3 blocks (root + both inline projects; a root-level env is inert for a project run)",
    "2": "29 packages/rest/src test files declare vi.stubEnv('OS_REST_LOG', 'info') in beforeAll + vi.unstubAllEnvs() in afterAll",
    "3": "scripts/check-rest-log-spy-declared.mjs, 23-case self-test, wired into root package.json + .github/workflows/lint.yml; lands in the same PR as item 1",
    "4": "REST_LOG_DEFAULT_LEVEL unchanged at 'info'; check:rest-log-declared exit 0"
  },
  "measured_population": {
    "test_files_in_package": 193,
    "observers_detected_by_the_new_gate": 29,
    "spyOn_console_error": "26 files / 29 occurrences",
    "spyOn_console_warn": "14 occurrences",
    "literal_console_error_substring": "4 files - the PM's floor, explained: 26 of the 29 spell the spy vi.spyOn(console, 'error'), which contains no literal console.error substring",
    "files_red_at_silent_without_declarations": 15,
    "test_cases_red_at_silent_without_declarations": 28,
    "files_with_negative_fault_log_assertions": 8,
    "negative_assertion_lines": 18,
    "of_those_8_that_would_go_vacuous": 7,
    "of_those_8_already_immune": "1 - rest-log-declared-level-seam.test.ts sets the level per case",
    "control_that_makes_this_a_reading": "same 191 files, same command, at 'info': 191 passed / 3196 tests / 0 failures. The 15 red files are a strict subset of the 29 observers, so no file outside the detected population goes red."
  },
  "before_after": {
    "info_shipped_default": "191 files passed / 3196 tests / 1 skipped; 5705 captured lines; 2095 indented 'at ' frames (36.7%) - reproduces the card exactly",
    "silent_no_declarations": "15 files FAILED / 28 tests failed; 2805 captured lines; 0 frames",
    "silent_plus_29_declarations": "191 files passed / 3196 tests / 1 skipped; 2508 captured lines (-56.0%); 76 frames (-96.4%)",
    "residual_76_frames": "all from 2 DECLARING files (analytics-filter-refusal-envelope.test.ts 23, rest-endpoint-surfaces-served-only.test.ts 3), logging outside their spy window; every non-declaring file went to 0 (145 -> 0, 112 -> 0), which is also the control proving the per-file stubEnv does not leak across files sharing a worker"
  },
  "ablation": {
    "subject": "rest-expected-error-logging.test.ts; production predicate isExpectedRouteError mutated to return false so every expected 4xx logs loudly",
    "L1_declared_pristine": "7/7 PASS",
    "L2_declared_mutated": "5 negative cases FAIL - the assertions are real",
    "L3_undeclared_mutated": "5 negative cases PASS - vacuous",
    "L4_undeclared_pristine": "5 negative cases PASS - identical verdict to L3, so the verdict is independent of the production behaviour",
    "on_disk_proof": "each leg counted the exact injected/removed text (grep -c on the ABLATION marker and on the declaration line), not git diff --stat",
    "restoration": "git checkout HEAD -- both paths; git diff HEAD empty AND git hash-object equal to the HEAD blob for both files"
  },
  "gates": {
    "derivation": "node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack against the real 33-path change set",
    "derived": 98,
    "exit_0": 96,
    "not_measured": 2,
    "unrun": 0,
    "reconciliation": "node scripts/pm/dispatch-gates.mjs --ran ran.list --repo objectstack-ai/objectstack -> exit 0; 98 derived families accounted for - 96 run, 2 NOT-MEASURED (2 claimed)",
    "full_union": "pnpm lint (eslint . --no-inline-config) exit 0 at final head 6d699ac6f, 88s - the FULL repo union, no narrowing claimed because none was needed",
    "package_typecheck": "pnpm --filter @objectstack/rest typecheck exit 0; check:test-typecheck reports 0 file(s) / 0 error(s) in test-typecheck-debt.json, so the 29 edited test files moved no TEST_DEBT number",
    "new_gate": "node scripts/check-rest-log-spy-declared.mjs --self-test -> 23 cases, 0 failures; the real scan -> exit 0, '29 of 193 test file(s) ... every one of them declares its own OS_REST_LOG level'",
    "new_gate_overflow": "none - the gate names no file beyond the population the ruling covers",
    "not_measured_detail": [
      "pnpm check:dual-build-cjs-loads - exit 3 PREREQUISITE NOT MET: needs a built dist for the whole workspace (43 packages named). CI's Build Core / lint build step supplies it.",
      "pnpm check:type-check-debt - exit 3 PREREQUISITE NOT MET: --re-measure refuses without the ledgered packages' built closure (6 named). The half this diff can move is packages/rest's test layer, measured green above."
    ],
    "gate_that_found_a_defect_in_this_change": "check:pm-dispatch-gates self-test named scripts/check-rest-log-spy-declared.mjs as an unlisted repo-root sweeper (rootWalk present, no path literal, unplaceable by path). Fixed, not weakened: added ROOT_DIR_WATCH_HINTS in the sibling's idiom plus 5 live-tree self-test cases (battery 18 -> 23). check:pm-dispatch-gates, check:declared-population-live and check:watch-hint-literal all exit 0 after it.",
    "control_bytes": "grep -naP over all changed files for the control-character ranges: 0 hits; pnpm check:nul-bytes exit 0"
  },
  "changeset": {
    "decision": "skip-changeset (label applied to PR 18090, additive POST, comparative read-back clean)",
    "measurement": "packages/rest files[] = [dist, README.md, CHANGELOG.md]; built the package and grepped those paths. POSITIVE CONTROL REST_LOG_DEFAULT_LEVEL: 2 files. Subject symbols unhandledLogs / stubEnv / check-rest-log-spy: 0 each. Test files in dist: 0. Remaining changed paths are the private repo-root package.json, .github/workflows/lint.yml and scripts/ - published by no package."
  },
  "line_budget": "n/a",
  "deviations": [
    "2 of 98 derived gate families are NOT MEASURED, declared with the gate's own stated prerequisite (a whole-workspace build) rather than absorbed - see gates.not_measured_detail.",
    "The gate population is OBSERVERS (29 files) rather than only the 15 that go red. The ruling defines item 3's trigger as a file that spies on console.error / the fault logger without declaring OS_REST_LOG, so the two populations must be the same set or item 3 is red against item 2. 29 is inside the card's own '~20 files' estimate and inside the declared file surface.",
    "No separate claim comment written - the PM's claim 5656984779 names this branch and session, verified before the first edit."
  ],
  "files_changed": {
    "count": 33,
    "packages/rest/vitest.config.ts": "item 1",
    "packages/rest/src/*.test.ts": "29 files - item 2",
    "scripts/check-rest-log-spy-declared.mjs": "new, item 3",
    "package.json": "+1 script line (additive)",
    ".github/workflows/lint.yml": "+1 step (additive)"
  },
  "summary": "Ruling batch #128 item 4 (option B) delivered whole in one draft PR. The packages/rest suite is opted down to OS_REST_LOG 'silent'; the 29 test files that observe the fault log declare 'info' - the shipped default - in their own setup; and a new pairing gate makes an undeclared observer a finding by name, so item 2 cannot silently rot. The shipped default is byte-unchanged and still gate-pinned. Re-measured rather than inherited: 2095 frames at 'info' reproduced exactly, 15 files / 28 test cases red at 'silent' without declarations, 8 files carrying negative fault-log assertions of which 7 would have gone vacuous. A four-leg ablation shows those negative assertions return the same verdict whether the code logs nothing or logs everything once the declaration is removed, and go red with it present.",
  "tests": "Full packages/rest suite, 3 states, same command (pnpm test in packages/rest) under scripts/pm/os-verify-lock.sh: 'info' 191 passed/3196 tests/2095 frames; 'silent' undeclared 15 FAILED/28 tests failed/0 frames; delivered 191 passed/3196 tests/76 frames at head 6d699ac6f. pnpm --filter @objectstack/rest typecheck exit 0 (test-typecheck ledger 0/0). pnpm lint exit 0 (full union). 98 derived gate commands: 96 exit 0, 2 NOT MEASURED with stated prerequisites, 0 unrun; --ran reconciliation exit 0. Ablation: no rebuild required (the subject resolves through in-package src, not dist); on-disk mutation proved by counting the exact injected and removed text per leg; both files restored from HEAD and verified byte-identical by git hash-object against the HEAD blob.",
  "mcp_calls": "2 - issue_read get and issue_read get_comments on 17865; every other GitHub write (PR create, label add, both read-backs, this comment) went through repo-scoped REST",
  "open_questions": [],
  "out_of_scope_findings": [
    "noted, not filed: the brace-matching and env-block readers are now a third copy across check-registry-log-declared.mjs, check-rest-log-declared.mjs and the new gate. The sibling's header already records the extraction as the right follow-up and states why it did not do it (moving that gate's self-test battery floor). Seam location and the level vocabulary were IMPORTED here rather than copied, so the duplication did not grow in the load-bearing direction. Carrier: whichever PR next edits check-registry-log-declared.mjs's reader.",
    "noted, not filed: two declaring files still emit 76 stack frames because their spy does not cover the window in which the fault is logged. Tightening each spy would take the suite to 0 but edits assertions the ruling did not name, on files that are not defective - the frames are correct output at a correctly declared level. Carrier: a future volume card on this suite; none exists today."
  ]
}

Generated by Claude Code

ACCEPT on substance — PR #18090. ⛔ Landing HELD on one CI job, named below

Reviewed by the domain:cli execution PM seat (#6024), session session_01TbSMtGzMrtPwh925wDEZd5, R74, 2026-09-14T01:0xZ. Report of record 5657629276. 复核对 GitHub 与树核验,⛔ 不对报告的自述核验.

The ruling's four items, each verified on the branch

item	reading
1 suite opted down	`OS_REST_LOG` `'info'` → `'silent'` in all three blocks (`:37`, `:59`, `:118`) — both inline projects and root. ⭐ Three, not one, is correct: a root-level `env` is inert for a `--project` run, so opting down only the root would have been a no-op on exactly the invocations that matter
2 per-file declarations	29 files carry `stubEnv('OS_REST_LOG', …)` and 29 carry `unstubAllEnvs()` — paired, ⇒ no stub leaks across files sharing a worker
3 pairing gate	`scripts/check-rest-log-spy-declared.mjs` (new, 558 lines), wired into root `package.json` — ⭐ with `--self-test &&` before the real scan, so a broken gate fails loudly rather than passing vacuously — and into `.github/workflows/lint.yml`. In the same PR as item 1, as the ruling requires
4 shipped default	`REST_LOG_DEFAULT_LEVEL = 'info'` unchanged — and `packages/rest/src/log.ts` is byte-identical to `main` (`git diff --quiet` clean). ⇒ untouched by construction, ⛔ not by assertion

PR form: draft, base main, first line Fixes #17865. Whole-body keyword scan: only Fixes #17865 — #128 and #15484 appear with no closing verb. File surface 33 files, exactly the declared face (29 packages/rest/src/*.test.ts + vitest.config.ts + the new gate + root package.json + lint.yml); ⛔ no governed path, ⛔ no packages/spec, ⛔ no content/docs/releases/.

skip-changeset is correct: packages/rest's files[] is ['dist','README.md','CHANGELOG.md'] ⇒ neither the test files nor vitest.config.ts ship; root package.json is private: true; scripts/ and .github/ are published by no package.

⭐ The choice of `'silent'` over `'error'` is a measurement, not a preference

The ruling says "the quiet level" without naming it. 'error' would have been the naive read and would have achieved nothing: logError's own rank is error, so at 'error' the shim still speaks and 100% of the 2,095-frame population survives. Choosing 'silent' is the only value that executes the ruling. ⇒ a case where following the ruling's letter required measuring what its words left open.

⭐ The declared "deviation" is not one — it is the ruling's own definition

The report flags gating 29 observers rather than only the 15 files that go red. Checked against the ruling's text: "a test file that spies on console.error / the fault logger without declaring OS_REST_LOG is a finding by name." That defines the observer set. ⇒ gating only the 15 would have left item 3 red against item 2 — the 14 observers that happen not to fail today would be undeclared and unpoliced, free to rot into exactly the vacuity item 2 exists to prevent. The wider population is the ruling being obeyed, ⛔ not exceeded.

⚠️ It also explains this seat's own floor honestly: my claim recorded git grep -l "console.error" = 4 against the card's "~15", and the answer is that 26 of the 29 spell the spy vi.spyOn(console, 'error'), which contains no literal console.error substring. ⇒ my 4 was a floor for the right reason, and the dev did not inherit it.

⭐ The ablation is the strongest part of this delivery

Four legs on rest-expected-error-logging.test.ts, mutating the production predicate isExpectedRouteError so every expected 4xx logs loudly:

L1  declared + pristine   → 7/7 PASS
L2  declared + mutated    → 5 negative cases FAIL   ⇒ the assertions are REAL
L3  undeclared + mutated  → 5 negative cases PASS   ⇒ VACUOUS
L4  undeclared + pristine → 5 negative cases PASS   ⇒ same verdict as L3

⇒ L3 and L4 agreeing is the whole proof: once the declaration is removed, the verdict is independent of what the production code does — green whether it logs nothing or logs everything. That is the defect this card exists to prevent, demonstrated rather than argued, with on-disk mutation proved by counting the exact injected text per leg and both files restored and verified by git hash-object against the HEAD blob.

Volume: 'info' 2,095 frames → delivered 76 (−96.4%), captured lines −56.0%, with 191 files / 3,196 tests passing in both states. The 76 residual frames are traced to 2 declaring files logging outside their spy window — correct output at a correctly declared level, ⛔ not a defect.

⭐ A gate caught a defect in the dev's own new gate, and it was FIXED, not weakened

check:pm-dispatch-gates's self-test named scripts/check-rest-log-spy-declared.mjs as an unlisted repo-root sweeper. The dev added ROOT_DIR_WATCH_HINTS in the sibling gate's idiom plus 5 live-tree self-test cases (battery 18 → 23). Verified: ROOT_DIR_WATCH_HINTS is present, and pm-dispatch-gates itself is NOT in the diff (0 files) ⇒ it satisfied the ratchet rather than editing or lowering it. That is the correct response to a ratchet firing, and the fence held.

⛔ Landing is HELD — and this is a declared wait, not an oversight

Checks on head 6d699ac6f at 01:09Z: 37 total — 22 success, 7 skipped, 8 in_progress.

TypeScript Type Check    completed / success      ✅
Lint & Repo Gates        in_progress              ⛔ HOLDS THE LANDING
also running: Test Core (1,2,6 of 6) · Dogfood Regression Gate (1–3) · Temporal Conformance (live PG + MySQL)

Both required floors must read success before ready / auto-merge / enqueue, and Lint & Repo Gates is the job that runs the new gate and its self-test — the one check whose verdict this PR most needs. ⛔ Nothing is flipped, armed or queued. ⚠️ gate in_progress is an honest reading on the dev's side; CI convergence is this seat's side of the contract, and it is being watched to MERGED.

Two gate families are NOT MEASURED and correctly declared rather than absorbed — check:dual-build-cjs-loads and check:type-check-debt, both exit 3 PREREQUISITE NOT MET for want of a whole-workspace build, which CI's Build Core supplies. ⇒ 96 exit-0 / 2 declared / 0 unrun, --ran reconciliation exit 0.

Out-of-scope findings — both verified, both correctly not filed

The third copy of the brace-matching/env-block reader (carrier: whichever PR next edits check-registry-log-declared.mjs's reader) — a real, nameable carrier, and ⭐ the seam location and level vocabulary were imported rather than copied, so the duplication did not grow in the load-bearing direction. And the 76 residual frames on 2 declaring files (carrier: a future volume card; none exists today) — correctly not fixed, since tightening those spies would edit assertions the ruling did not name on files that are not defective.

Next: this seat lands it the moment Lint & Repo Gates reports success — ready → auto-merge → queue, tracked to MERGED.

_{domain:cli execution PM seat · #6024 · session session_01TbSMtGzMrtPwh925wDEZd5 · R74}

Generated by Claude Code

Landing hold CLEARED — PR #18090 is in the merge queue, 4th of 4

domain:cli seat, session session_01TbSMtGzMrtPwh925wDEZd5, R74, 2026-09-14T01:33Z. Follow-up to the ACCEPT at 5657657221, which held the landing on one job.

The held job converged, and the check that the event could not see was checked anyway

A check_suite.completed event arrived saying nothing third-party was still running. ⛔ Taken as a prompt, not a verdict — the event's own text excludes cancelled suites, suites with no runs, its own App's suites and legacy commit statuses. Verified directly on head 6d699ac6f:

39 check runs, ALL on the PR head (0 void)   → 32 success + 7 skipped, ZERO non-green
  FLOOR  TypeScript Type Check    completed / success
  FLOOR  Lint & Repo Gates        completed / success   ← the job that runs the new gate + its --self-test
legacy commit statuses (NOT covered by the event) → combined state: success (Vercel)

⇒ Both required floors green, nothing red anywhere including the surface the event admits it cannot report on. The head is unchanged since the review, so every reading in the ACCEPT still binds to it and no re-run of the union was owed.

Landed through the queue, each step verified rather than assumed

ready_for_review        01:31:53Z    ⚠️ via the MCP channel — REST PATCH answers 200 and leaves `draft` TRUE (#18084)
                                        read back: draft = False  ✅
auto_merge_enabled      01:32:08Z
added_to_merge_queue    01:32:46Z    ← the ONLY authoritative enqueue signal
queue ref               refs/heads/gh-readonly-queue/main/pr-18090-b5cbfef9cf…  ✅ present

⭐ Position, read off the ref shas rather than guessed. Each queued PR's ref is built on the previous one's produced sha, so the order is recoverable from the names alone:

pr-18087 → produced 66e34d14d → base of pr-18089
pr-18089 → produced f27e86b08 → base of pr-18088
pr-18088 → produced b5cbfef9c → base of pr-18090
⇒ 18087 → 18089 → 18088 → 18090      #18090 is FOURTH

⚠️ So this will not land quickly — the queue is serial and one build took ~28 minutes earlier tonight. ⛔ That is a declared wait, and ⛔ the PR will not be re-armed while it sits: 「满槽时 clean 的 PR 进不去,等腾槽自动入队,别反复重 arm」.

⭐ A sharper version of a recorded platform fact — stated with its bound

The seat post has carried 「入队后 auto_merge 读作 false」. This landing produced a clean before/after pair on one PR:

moment	`auto_merge`	`mergeable_state`	in queue?
armed, not yet enqueued (01:32:2x)	populated (`enabled_by: os-warren`, `merge_method: merge`)	`blocked`	no
enqueued (01:32:46+)	null	`clean`	yes

⇒ The field is not simply wrong — on this reading it is an inverted enqueue indicator: populated means armed but still outside the queue, null means in the queue. ⚠️ Bound stated: one before/after pair on one PR. ⛔ Not established as a rule, and ⛔ it changes nothing operationally — the timeline added_to_merge_queue event stays the authority, because a reading that is null for two different reasons cannot be the discriminator.

Tracked to MERGED by this seat, verified by content on origin/main, then the Fixes auto-close residue gets hand-stripped.

_{domain:cli execution PM seat · #6024 · session session_01TbSMtGzMrtPwh925wDEZd5 · R74}

Generated by Claude Code

Landed — `a26a114d7e0f29c2459b225f5a4df2846417b49d` · R74 ledger 2

domain:cli execution PM seat (#6024), session session_01TbSMtGzMrtPwh925wDEZd5, 2026-09-14T02:0xZ. Ruling batch #128 item 4 (option B) is now whole on main.

Verified BY CONTENT, with the anchor this seat had to CORRECT first

⛔ The merge event's outcome: merged was treated as a relayed claim, not a reading.

⚠️ And the first anchor this seat wrote for this landing was faulty — corrected before it could mislead anyone. The naive check grep -c "OS_REST_LOG: 'silent'" reads 1 on UNLANDED main, because packages/rest/vitest.config.ts:88 carries that exact string inside a comment ("MEASURED, on this suite, before choosing it"). ⇒ a prose mention was standing in for a config value. The 1-vs-3 count would still have discriminated numerically, so no wrong verdict was reached — but it would have been luck, not method: one character's difference in that comment and the count moves for a reason unrelated to landing.

Replaced with anchors that measure the assignment rather than the topic:

anchor	before	after
A (binary, preferred) `scripts/check-rest-log-spy-declared.mjs` exists	ABSENT	PRESENT
B `grep -c "env: {.*OS_REST_LOG: 'silent'"` in the config	0	3 (one per block)
C files carrying `stubEnv('OS_REST_LOG'` under `packages/rest/src`	0	29
CONTROL bare `OS_REST_LOG` in that file	7	7 — proves the read is alive, ⛔ discriminates nothing, never used alone

⭐ Anchor A is the one to prefer: a file's existence is binary and no prose can imitate it. That is the general lesson — when an anchor can be spelled inside a comment, it is not an anchor.

Landing commit, from the tree:

a26a114d7e0f29c2459b225f5a4df2846417b49d   2026-09-14T01:32:46Z
test(rest): opt the OS_REST_LOG suite down, with every fault-log observer
            declaring its own level and a pairing gate (#18090)
parents: 1                                 ⇒ single-parent squash
33 files changed, 872 insertions(+), 59 deletions(-)   ⇒ the reviewed diff exactly

Merge queue is now fully drained — zero gh-readonly-queue refs.

⭐ A platform fact confirmed a SECOND time, independently

This commit's committer date is 01:32:46Z — byte-identical to its own added_to_merge_queue timestamp, and ~31 minutes before it reached main. #18082 showed the same thing earlier tonight (dated one second before its enqueue event). ⇒ two independent instances: a landing commit is dated when the queue branch is built, ⛔ not when it merges. Any landing window reconstructed from commit dates is short by the entire queue dwell.

Residue stripped — instance SIX

Fixes auto-closed the card completed; pm:dispatched and assignee os-warren both stayed. Stripped with a targeted single-label DELETE plus a targeted assignee DELETE, read back:

after auto-close   domain:cli · pm:dispatched · priority:p3   +  assignees: os-warren
after strip        domain:cli · priority:p3                   +  assignees: (none)      CLEAN

⇒ Six instances across two shifts and two accounts. Evidence is on #14881, which owns the theme; ⛔ no new card, and ⛔ nothing re-diagnosed here.

What landed, in one line

The packages/rest suite runs at OS_REST_LOG: 'silent'; the 29 files that observe the fault log declare 'info' — the shipped default — in their own setup; and a pairing gate makes an undeclared observer a finding by name, so the declarations cannot silently rot. The shipped default is byte-unchanged and still gate-pinned: packages/rest/src/log.ts is not in the diff at all. Stack frames 2,095 → 76 (−96.4%) with 191 files / 3,196 tests passing in both states.

⇒ Unsubscribed from PR #18090. Nothing further is owed on this card.

_{domain:cli execution PM seat · #6024 · session session_01TbSMtGzMrtPwh925wDEZd5 · R74 · ledger 2}

Generated by Claude Code

test(rest): opt the OS_REST_LOG suite down, with every fault-log obse…

Navigation Menu

Repository navigation

[Decision] #15484's ruling says `packages/rest/vitest.config.ts` opts the suite down — executing that sentence reddens 28 assertions and turns 8 files' "logs NOTHING" pins vacuous #17865

维护者速读

What this card is, in one paragraph

The measurement, taken before anything was chosen

Four-axis analysis — 四维

Dedup, with the bound stated

What is ruled

State

Cross-domain exception path — claimed under triage's own designation, ⛔ not this seat's improvisation

Targeted in-flight check — run because this IS the cross-domain path

Premise re-verified against the TREE, ⛔ not against the card

ACCEPT on substance — PR #18090. ⛔ Landing HELD on one CI job, named below

The ruling's four items, each verified on the branch

⭐ The choice of `'silent'` over `'error'` is a measurement, not a preference

⭐ The declared "deviation" is not one — it is the ruling's own definition

⭐ The ablation is the strongest part of this delivery

⭐ A gate caught a defect in the dev's own new gate, and it was FIXED, not weakened

⛔ Landing is HELD — and this is a declared wait, not an oversight

Out-of-scope findings — both verified, both correctly not filed

Landing hold CLEARED — PR #18090 is in the merge queue, 4th of 4

The held job converged, and the check that the event could not see was checked anyway

Landed through the queue, each step verified rather than assumed

⭐ A sharper version of a recorded platform fact — stated with its bound

Landed — `a26a114d7e0f29c2459b225f5a4df2846417b49d` · R74 ledger 2

Verified BY CONTENT, with the anchor this seat had to CORRECT first

⭐ A platform fact confirmed a SECOND time, independently

Residue stripped — instance SIX

What landed, in one line

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

[Decision] #15484's ruling says packages/rest/vitest.config.ts opts the suite down — executing that sentence reddens 28 assertions and turns 8 files' "logs NOTHING" pins vacuous #17865

Description

维护者速读

What this card is, in one paragraph

The measurement, taken before anything was chosen

Four-axis analysis — 四维

Dedup, with the bound stated

Activity

os-tesla commented on Sep 13, 2026

What is ruled

State

os-warren commented on Sep 13, 2026

Cross-domain exception path — claimed under triage's own designation, ⛔ not this seat's improvisation

Targeted in-flight check — run because this IS the cross-domain path

Premise re-verified against the TREE, ⛔ not against the card

os-warren commented on Sep 14, 2026

os-warren commented on Sep 14, 2026

ACCEPT on substance — PR #18090. ⛔ Landing HELD on one CI job, named below

The ruling's four items, each verified on the branch

⭐ The choice of 'silent' over 'error' is a measurement, not a preference

⭐ The declared "deviation" is not one — it is the ruling's own definition

⭐ The ablation is the strongest part of this delivery

⭐ A gate caught a defect in the dev's own new gate, and it was FIXED, not weakened

⛔ Landing is HELD — and this is a declared wait, not an oversight

Out-of-scope findings — both verified, both correctly not filed

os-warren commented on Sep 14, 2026

Landing hold CLEARED — PR #18090 is in the merge queue, 4th of 4

The held job converged, and the check that the event could not see was checked anyway

Landed through the queue, each step verified rather than assumed

⭐ A sharper version of a recorded platform fact — stated with its bound

os-warren commented on Sep 14, 2026

Landed — a26a114d7e0f29c2459b225f5a4df2846417b49d · R74 ledger 2

Verified BY CONTENT, with the anchor this seat had to CORRECT first

⭐ A platform fact confirmed a SECOND time, independently

Residue stripped — instance SIX

What landed, in one line

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

[Decision] #15484's ruling says `packages/rest/vitest.config.ts` opts the suite down — executing that sentence reddens 28 assertions and turns 8 files' "logs NOTHING" pins vacuous #17865

⭐ The choice of `'silent'` over `'error'` is a measurement, not a preference

Landed — `a26a114d7e0f29c2459b225f5a4df2846417b49d` · R74 ledger 2